AMD and Hynix brings massive improvement over GDDR5
AMD announced that it would be working with memory manufacturer SK Hynix for a joint development of High Bandwidth 3D Stacked memory. This marks that the memory technology took one step closer to full commercialization, and AMD has indicated that the first application for the High Bandwidth Memory (HBM) would be GPUs.
High Bandwidth Memory (HBM)
Let’s start with some figures. Hynix suggests that they can achieve an incredible 65% performance improvement over GDDR5, 40% power reduction, and a size reduction of 37 times over DDR4. While these figures are not fixed, there would be a substantial increase in hashing performance with greater power savings, which are what miners were looking for.
HBM does not increase bandwidth by increasing memory frequency as in GDDR5. It is merely done by increasing the channels which the data can travel. Since GDDR5 has hit a significant ceiling where neither design nor material change would improve the speed significantly, HBM is made by essentially stacking multiple memory chips on top of each other with the die-stacking technology, increasing the number of memory floors to have a 3D memory structure.
The conventional RAM that utilizes GDDR5 is much faster in terms of frequency, but HBM can move more data in the same time period of GDDR5. The 3 dimensional memory array of HBM makes data storage more compact, and hence retrieval became more efficient. This also brings the benefit of lower power consumption.
AMD will possibly use 3D stacked memory in their next generation 20nm GPUs likely to be codenamed ‘Pirate Islands’. This has not been confirmed but such bandwidth makes a lot more sense when it comes to GPUs. With both HBM and ASIC poised to be released later this year, miners will decide which will be the better choice for scrypt mining.